A checklist does not change how careful a person is. It changes which of the steps they already know survive a busy day, because a busy day is where known steps get skipped. The evidence behind that claim is thinner than the books suggest: it rests mostly on before-and-after hospital studies, one mandate study found nothing, and the strongest result in intensive care was really a checklist plus a change in who was allowed to stop the procedure. What carries over to a three to twenty person service firm is the mechanism. Where the steps are known and get skipped under pressure, a list helps, and where the work is judgment, it does not.
Where the checklist came from
The first checklist in the modern sense came out of a crash. On 30 October 1935 the Boeing Model 299 stalled and crashed at Wright Field because Major Ployer Peter Hill had not released the gust lock, and the Smithsonian's account of the event notes the aircraft required at least 30 steps to prepare it for takeoff. The response was not more training. Gawande's account in The New Yorker in 2007 makes the point: it was hard to imagine more expertise than Hill's, so the test pilots created a pilot's checklist, with step-by-step checks for takeoff, flight, landing, and taxiing. The line in that account transfers: the plane was too complicated to be left to the memory of any pilot, however expert.
In aviation the list is not read silently. It is a challenge-and-response: one person calls the item, another has to answer, and a mismatch is caught in the moment. Checklists in aviation fail the same way they fail in an office. Reviewing the accident record, safety investigators found distraction and interruption the most commonly cited problem, with crews stopping mid-check and resuming uncertain where they had left off, and some crews so familiar with the procedure that they did not even notice their own errors.
Exhibit 1
A list catches the steps that are known and get skipped under pressure, and leaves the judgment calls untouched
What the surgical evidence shows
In the study Atul Gawande and colleagues published in 2009, eight hospitals in eight countries ran the same 19-item surgical safety checklist, and the rate of death fell from 1.5% before the checklist to 0.8% after, while inpatient complications fell from 11.0% to 7.0%. Haynes and colleagues describe that fall as a 36% drop in postoperative complication rates, with death rates falling by a similar amount. What actually changed in those operating rooms was adherence: six basic safety steps were all completed for 34.2% of patients before the checklist and 56.7% after. The checklist also required a formal pause in care during surgery for preoperative team introductions and briefings and postoperative debriefings, which is a design detail worth keeping, because the list only ran at a point where the room stopped.
That study is not a randomised trial. The same operating rooms were measured before and after, and the authors say they chose that design because they could not randomise rooms without significant cross-contamination. Haynes and colleagues raise the Hawthorne effect against their own result, writing that the contribution of the Hawthorne effect is difficult to disentangle in this study, and they kept the whole study under a year to limit confounding by secular trends. None of that makes the number wrong. It makes it a before-and-after number, and it says what the number can and cannot carry when you move it out of the operating room.
What the wider evidence base says
Bergs and colleagues, in a 2014 review, pooled the seven studies that met their criteria and reported a risk ratio of 0.59 for any complication, but wrote that the evidence cannot be regarded as definitive in the absence of higher-quality studies, and flagged marked methodological heterogeneity among studies. The one randomised test in this evidence base is Haugen and colleagues' stepped-wedge trial in two Norwegian hospitals, which compared 2,212 control procedures with 2,263 checklist procedures and found complications falling from 19.9% to 11.5%, while the mortality reduction was significant in one hospital and not across both. The strongest results in the base come from before-and-after studies, the mandate study found no improvement, and the meta-analysis calls its own evidence suggestive rather than definitive.
The Michigan intensive care work
Peter Pronovost's Michigan project reported the median rate of catheter-related bloodstream infection falling from 2.7 per 1,000 catheter-days at baseline to 0 at three months, and the mean rate from 7.7 to 1.4 at 16 to 18 months, across 103 reporting intensive care units. The checklist itself was plain and known: hand washing, full barrier precautions, cleaning the skin with chlorhexidine, avoiding the femoral site where possible, and removing catheters that were no longer needed.
The part that matters is what the checklist was not. The team's own follow-up describes five specific CDC recommendations coupled with a comprehensive patient safety intervention to improve culture, educate staff, learn from mistakes, and involve senior leaders. The intervention was a statewide safety culture programme, and a checklist hung inside it. The list was the mechanism and the culture ran it, which is the part a small firm can actually copy.
What the Ontario mandate found
When Ontario made surgical checklists universal, the result went the other way. Across 101 hospitals and roughly 109,000 procedures before and 106,000 after, adjusted mortality moved from 0.71% to 0.65% and complications from 3.86% to 3.82%, neither one a significant change. David Urbach and colleagues concluded in 2014 that checklist implementation in Ontario was not associated with significant reductions in operative mortality or complications.
Exhibit 2
The 2009 study ran the list at a formal pause and measured whether it was used, and the 2014 mandate only made it universal
The mandate did not reproduce the result. That is all the study supports: it measured administrative data over three-month windows either side of adoption, and it did not measure whether the checklists were actually used. A list handed down without the culture around it is paper.
What this means for a service firm
In a firm of three to twenty, the work that behaves like the operating room is the recurring, multi-step, interruptible work: onboarding a client, handing a job between two people, month-end close, publishing anything, sending a quote. Those are steps that are known in advance, and they are the ones that get skipped when the day gets loud. Interruption is the specific mechanism, the same one the research on context switching describes for the owner who does everything, and the step that gets skipped is usually the one that was never written down.
The work that does not behave like the operating room is the advice itself. The judgment call on which engagement to take, what to tell the client about the budget, whether the quote is the right shape: no list runs that. And the evidence base does not either. Every row in it is about known steps under load: releasing a gust lock, giving antibiotics within sixty minutes, washing hands, confirming a patient's name. No study in it tested judgment. The claim that checklists do least for judgment is an argument from the absence of evidence, and that is how it should be read.
Exhibit 3
The candidates are the five recurring jobs that already repeat on a known sequence of steps
The candidates for a small practice: onboarding a client, which has a checklist of its own worth borrowing; handing a job between two people; the month-end close; publishing; and the quote going out. Each runs on a fixed sequence of steps the people in the firm already know. That is the profile the evidence actually speaks to.
How to write one that survives
The design rules come from a short guide, "A Checklist for Checklists," that Gawande and Boeing's Dan Boorman wrote. They set the test for putting an item on a list at all: it has to be a critical safety step and in great danger of being missed, and not adequately checked by other mechanisms. The same guide says a checklist should use natural breaks in workflow (pause points), keep fewer than 10 items per pause point, fit on one page, and carry a visible date of creation or revision. It also prints a limit: a checklist is not a teaching tool or an algorithm.
Keep it short and put it at a pause point, at the moment the work changes hands or changes state, not in the middle of a task. If the people running it are experienced, the list works confirm-afterwards: they do the work, then confirm the steps against the list. If they are not, it works read-as-you-go, the way the aircraft list runs, with the steps called out as they happen. The person who does the work owns the list, and the list is revised when it fails, with the revision date on the page.
There is a trap that kills most of them: a checklist that grows a line every time something goes wrong until nobody reads it. The same discipline that keeps a list at one page and under ten items is what keeps the weekly review at 30 minutes and the small routines of a steadier week running at all. The checklist is one of those routines and earns its place the same way: short, at a fixed point, owned by a person, cut back the moment it stops being read.
Where we work on this in a client engagement is the management consulting practice.